A Congestion Control Method Of SDN Data Center Based On Deep Q-learning

As a future network architecture,the SDN(Software Defined Network)has been widely adopted by the Data Center Network(DCN).With the development of the big data and cloud computing,the number of nodes and the number of flows in the SDN data center are increasing,and the data center is facing the risk of network congestion.Deep Q Network(DQN)is one of the classic algorithms in Deep Reinforcement Learning(DRL).The algorithm combines the advantages of both deep learning and reinforcement learning to solve the high-dimensional raw input and decision control problems.The introduction of new artificial intelligence(AI)technology to solve the congestion problem,thereby improving the overall performance of the network,is a congestion control method adapted to the SDN data center network in line with the development trend of the network intelligence.This thesis studies the congestion control problem of data center based on SDN.The congestion control method has three characteristics.First,given the characteristics of SDN,we use a congestion control method based on flow.Secondly,we introduce deep reinforcement learning techniques to provide intelligence for congestion control method in SDN data center.Finally,we allocate the rate globally for the entire network through the controller,which not only avoids congestion in the entire network,but also makes the data link utilization of the network as high as possible,thereby achieving congestion control of the entire data center.Based on the existing congestion control algorithm of SDN data center based on reinforcement learning,this thesis proposes a congestion control algorithm of SDN data center based on Sarsa(?).In order to overcome the weakness of reinforcement learning without high-dimensional sensing ability,this thesis introduces deep reinforcement learning and proposes a congestion control algorithm of SDN data center based on DQN.We made a congestion control comparison test on the proposed congestion control methods of SDN data center based on Sarsa(X)and DQN,the congestion control method based on Sarsa,and the traditional congestion control method based on On-demand.The test results show that the congestion control methods can be effectively performed based on Sarsa,Sarsa(?)and DQN.The test results also show that the congestion control method based on DQN is superior to the other three methods in performance parameters such as algorithm convergence speed,link utilization rate and flow rate allocation percentage.